Leveraging Large Language Models for End-User Website Generation

Abstract

This work introduces an innovative approach that harnesses the power of large language models (LLMs) to facilitate the creation of websites by end users through natural language specifications. Our key contribution lies in a user-oriented method that utilizes prompt engineering, compelling the LLM response to adhere to a specific template, which in turn enables direct parsing of the model’s responses, allowing users to focus on refining the generated website without concerning themselves with the underlying code. The engineered prompt ensures model efficiency by implementing a modification strategy that preserves context and tokens generated in the LLM responses, updating only specific parts of the code rather than rewriting the entire document, thereby minimizing unnecessary code revisions. Moreover, our approach empowers LLMs to generate multiple documents, augmenting the user experience. We showcase a proof-of-concept implementation where users submit textual descriptions of their desired website features, prompting the LLM to produce corresponding HTML and CSS code. This paper underscores the potential of our approach to democratize web development and enhance its accessibility for non-technical users. Future research will focus on conducting user studies to ascertain the efficacy of our method within existing low-code/no-code platforms, ultimately extending its benefits to a broader audience.